Event-based uncertainty assessment of sediment modeling in a data-scarce catchment

Abstract

Uncertainties in watershed modeling need to be addressed to increase predictive accuracy and enhance model interpretation. Nevertheless, event-based characteristics incorporating management concerns were rarely considered in the sediment modeling within an uncertainty assessment framework, which might lead to biased decision makings in watershed management. In this study, the event-based likelihood measure was developed to improve the understanding and prediction of the sediment dynamics in a data-scarce catchment in the Three Gorge Reservoir Region (TGRR), China. The uncertainty assessment is based on the Generalized Likelihood Uncertainty Estimation (GLUE) approach applied to the Hydrological Simulation Program - Fortran (HSPF) model. The impact of the Nash-Sutcliffe efficiency (NSE) as the traditionally used likelihood measure was also investigated as a comparison. The results showed that the event-based likelihood measure had advantages in judging critical parameters. Sensitive parameters in the soil erosion and channel transport and their impacts on different phrases of sedimentograph were recognized. The event-based likelihood measure was much more discriminating by efficiently eliminating unsatisfactory parameter sets. Better validity in efficiently reducing the parameter uncertainty and a higher identifiability was achieved by the newly developed likelihood measure. The event-based likelihood measure resulted in more progressive reductions in predictive uncertainty in terms of uncertainty bounds and evaluation criteria. The proposed uncertainty assessment framework yields improved predictions and an increased understanding of sediment response behaviors in a data-limited environment.